Matlab

Two principal component analysis (PCA) models⁷⁵ were developed including EPS spectra from the two E. coli and two S. Enteritidis strains, in single- and dual-species biofilms. Prior modeling, spectra were pre-processed with standard normal variate (SNV) followed by the application of a Savitzky-Golay filter (15 smoothing points, 2^nd order polynomial, and 1^st derivative)^{76 ,77} and mean-centered. All data analyses were performed in Matlab version 7.9 (Mathworks, USA) and the PLS Toolbox version 5.5.1 for Matlab (Eigenvector Research, USA).

Calculations and data processing were
performed using Microsoft Excel, MATLAB and PLS_Toolbox (Eigenvector
Research, Inc.). Kinetic simulations were performed using COPASI.

Due to the large amount of spectral data, the 5 spectra of each sample were averaged before data analysis. The mean spectra were pre-processed with standard normal variate (SNV) and Savitzky-Golay filter (15 smoothing points, 2nd order polynomial and 1st derivative) [33 (link)] to remove baseline drifts and further mean centered. Other data pre-treatments were tested as: (I) different combinations of SNV and SavGol filter (SNV + mean center; SavGol + mean center); (II) different windows of the SavGol filter (9–15) and also the second derivative; (III) multiplicative scatter correction (MSC) and (IV) autoscale. It should be stressed that the best results were obtained with the above-mentioned pre-treatment. Spectra were further modelled by Principal component analysis (PCA) [34 ]. Outliers were verified by Q Residuals versus Hotelling T^2. The root mean square errors of calibration (RMSEC) and cross validation (RMSECV) of all the PCA models developed in the current study were presented in Table S3 (Supplementary Materials). All chemometric models were performed in Matlab version 9.5 Release 2018b (MathWorks) and PLS Toolbox version 8.7 (2019) for Matlab (Eigenvector Research, Manson, WA, USA).

The number of times the experiment was repeated is equal to the number of Raman spectra taken for statistics to calculate the average and SD. The typical number for each cancer type was minimum 1600 Raman spectra per image times the number of patients that gives around 16,000 Raman spectra for statistics. The data was expressed as a mean value ± standard deviation (SD). All data were analyzed using Origin software with the implemented one-way ANOVA using the Tukey test to calculate the value significance. p-Values ≤ 0.05 were accepted as statistically significant. A partial least squares discriminant analysis (PLSDA) with receiver operating characteristic area under the curve (ROC AUC) was carried out with MATLAB and the PLS_Toolbox (Eigenvector Research). The PLSDA was applied in a cross-validation scheme of venetian blinds with three splits.

A baseline correction was automatically applied to spectra within the instrument software prior to exportation during acquisition. All spectra were area-normalized before analysis using MATLAB (Mathworks). Chemometric analysis of acquired spectra was done in MATLAB equipped with PLS_Toolbox 9.0 (Eigenvector Research, Inc., Manson, WA). For Partial Least Squares Discriminant Analysis (PLS-DA), 100% calibration-cross validation models were employed; latent variables (LVs) were reported in corresponding tables and chosen lowest root-mean-square error of cross-validation (RMSECV) scores and classification error averages from both calibration and cross-validation, Fig. S1.† The preprocessing for each PLS-DA model, determined through a model optimizer, revealed that the most consistent and highest accuracy was achieved when employing “none,” indicating no preprocessing as the optimal model. Kruskal–Wallis analysis of variance tests (henceforth ANOVA) were utilized to rank specific band height differences between the averaged spectra of each sex.

Prior to multivariate modeling, NMR^{52 (link)} and MS spectral data were preprocessed (See SI Materials and Methods). Principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (oPLS-DA) models were generated to investigate the effect of K. brevis allelopathy on algal lipidomes (MATLAB and PLS Toolbox, version 7.9.1, Eigenvector research). Spectral features with discriminatory power in NMR-based models were annotated using the Human Metabolome Database and Chenomx Profiler while MS-based features were annotated using LOBSTAHS^{53 (link)}, KEGG, LIPID MAPS, and Metlin databases^{53 (link)–58 (link)}. Pooled extracts of each species were used to collect 2D NMR spectral data including: correlation spectroscopy (COSY), heteronuclear single quantum coherence (HSQC), and heteronuclear multiple bond correlation spectroscopy (HMBC) to aid in annotation.